![15/34 1. # include <stdio.h> //Standard Input/output library 2. # include <mpi.h> 3. int main(int argc, char** argv) 4. { 5. //MPI variables 6. int num_processes; 7. int curr_rank; 8. char proc_name[MPI_MAX_PROCESSOR_NAME]; 9. int proc_name_len; 10. //intialize MPI 11. MPI_Init(&argc, &argv); 12. //get the number of processes 13. MPI_Comm_size(MPI_COMM_WORLD, &num_processes); 14. 15. //Get the rank of the current process 16. MPI_Comm_rank(MPI_COMM_WORLD, &curr_rank); 17. // Get the processor name for the current thread 18. MPI_Get_processor_name(proc_name, &proc_name_len); 19. //Check that we're running this process. 20. printf("Calling process %d out of %d on %srn", curr_rank, num_processes, proc_name); 21. //Wait for all threads ot finish 22. MPI_Finalized(); 23. return 0; 24. } •Creates user specified dummy processes of equal size •Allocates the processes dynamically to each nodes •Displays the process number upon completion](https://image.slidesharecdn.com/presentation1spring2016finalcopy-200223190841/75/Senior-Design-Raspberry-Pi-Cluster-Computing-15-2048.jpg)
![ #include <stdio.h> #include <math.h> #include <mpi.h> #define TOTAL_ITERATIONS 10000 int main(int argc, char *argv[]) { //MPI variables int num_processes; int curr_rank; // keep track of the current for-loop iterations int total_iter; int step_iter; //variables used to calculate pi double pi; // the final value double curr_pi, h, sum, x; //step variables //start up MPI MPI_Init(&argc, &argv); MPI_Comm_size(MPI_COMM_WORLD, &num_processes); MPI_Comm_rank(MPI_COMM_WORLD, &curr_rank); //Iterate TOTAL_ITERATIONS to calculate PI within a certain error margin for(total_iter = 2; total_iter < TOTAL_ITERATIONS; total_iter++); { 16/34](https://image.slidesharecdn.com/presentation1spring2016finalcopy-200223190841/75/Senior-Design-Raspberry-Pi-Cluster-Computing-16-2048.jpg)
Senior Design: Raspberry Pi Cluster Computing
- 1. Final Presentation ELET4780 Michael Vistine Katy Rodriguez Ralph Walker 1/34
- 2. Michael Vistine Engineer 2/34 Katy Rodriguez Engineer RalphWalker Engineer
- 3. Motivation Hardware Design Description Software Design Improvements Timeline Current Status Immediate Tasks Conclusion Sources Questions?? 3/34
- 4. 4/34 Design • Cluster Computing •Compact • Active Cooling Raspberry Pi • Low Cost powerful device • Open Source Code Modifying Existing Design • Nodes vs. Performance • Wireless vs. Wired Performance • Add components for usability
- 5. 5/34 Photo courtesy of pcworld.com Pi1B+ Pi B 2 BeagleBone Processor 700 MHz 900-1000 MHz 1GHz Cores 1 4 1 RAM 512 MB 1 GB 512 MB Peripherals 4 USB Ports 4 USB Ports 2 USB Ports GPIO Capability 40 pins 40 pins 65 pins Memory Micro SD slot Micro SD slot 2 GB on board & Micro SD Price ~$30 ~$35 ~$55 Photo courtesy of ti.com Photo courtesy of adafruit.com
- 6. 6/34 Photo courtesy ofamazon • 5V per hub • Plug and play Insignia - 7-Port USB 2.0 Hub Photo courtesy of amazon Wireless Router TP-Link TL WR841N • 300Mbps wireless connection • Adjustable DHCP settings • Wireless On/Off switch • 4 LAN ports
- 7.
- 8. 8/34 Final Design • 3Dprinted using SolidWorks • Plexiglass • Wired/Wireless router • Heat Sinks and PC fans • Power hub
- 9. OPERATING SYSTEM– RAPSBIAN JESSIE(w/NOOBS) ◦ Easy to use ◦ Lightweight OS ◦ Open source ◦ Bash Terminal interface ◦ Linux/Unix kernel 9/34
- 10. Bash Terminal-used to: ◦ Edit and create files to manipulate the OS and ports i.e. setting up the host names and mounting drives ◦ Install software packages (i.e. openMPI, nfsserver) ◦ See IP addresses, Node settings and Network connections Style of syntax used to operate in Terminal: ◦ $ sudo apt-get install (“file”) – used to install files ◦ $ sudo nano (“file”) – used to edit files 10/34
- 11. OpenMPI: ◦ MessagePassing Interface used to implement parallel computing ◦ Takes the data and breaks it into smaller chunks and distributes it to the nodes to run simultaneously ◦ This method increases processing speed and efficiency ◦ Can compile and execute programs in C, C++, & Fortran ◦ GCC compiler is used to compile the program to be processed in a parallel fashion 11/34
- 12. First allpackages were updated ◦ Gfortran ◦ Nfs-common & Nfs-kernel-server ◦ Build-essential manpages-dev ◦ openmpi-bin/-doc libopenmpi-dev ◦ Etc. Go into the configurations using sudo raspi-config 12/34
- 13. Settings forthe master were the same as the slave nodes: ◦ Set the host names as rpi0 ◦ Enable ssh ◦ Overclock to “pi2” setting ◦ Set the memory split to 16 13/34
- 14. Install allthe same packages from the master node Sudo raspi-config to set all the same system preferences as the master node 14/34 Photo courtesy of www.raspberrypi.org
- 15. 15/34 1. # include<stdio.h> //Standard Input/output library 2. # include <mpi.h> 3. int main(int argc, char** argv) 4. { 5. //MPI variables 6. int num_processes; 7. int curr_rank; 8. char proc_name[MPI_MAX_PROCESSOR_NAME]; 9. int proc_name_len; 10. //intialize MPI 11. MPI_Init(&argc, &argv); 12. //get the number of processes 13. MPI_Comm_size(MPI_COMM_WORLD, &num_processes); 14. 15. //Get the rank of the current process 16. MPI_Comm_rank(MPI_COMM_WORLD, &curr_rank); 17. // Get the processor name for the current thread 18. MPI_Get_processor_name(proc_name, &proc_name_len); 19. //Check that we're running this process. 20. printf("Calling process %d out of %d on %srn", curr_rank, num_processes, proc_name); 21. //Wait for all threads ot finish 22. MPI_Finalized(); 23. return 0; 24. } •Creates user specified dummy processes of equal size •Allocates the processes dynamically to each nodes •Displays the process number upon completion
- 16. #include <stdio.h> #include <math.h> #include <mpi.h> #define TOTAL_ITERATIONS 10000 int main(int argc, char *argv[]) { //MPI variables int num_processes; int curr_rank; // keep track of the current for-loop iterations int total_iter; int step_iter; //variables used to calculate pi double pi; // the final value double curr_pi, h, sum, x; //step variables //start up MPI MPI_Init(&argc, &argv); MPI_Comm_size(MPI_COMM_WORLD, &num_processes); MPI_Comm_rank(MPI_COMM_WORLD, &curr_rank); //Iterate TOTAL_ITERATIONS to calculate PI within a certain error margin for(total_iter = 2; total_iter < TOTAL_ITERATIONS; total_iter++); { 16/34
- 17. //init sum sum = 0.0; //determine step size h = 1.0 / (double) total_iter; //the current process will perform operations on its rank //added by multiples of the total number of threads // rank = 3, for(step_iter = curr_rank +1; step_iter <= total_iter; step_iter += num_processes) { //determine the current step x = h * ((double) step_iter - 0.5); //add the current step values sum += (4.0/(1.0 + x * x)); } // resolve the sum into calculated value of pi curr_pi = h * sum; //reduce all processes' pi values to one value MPI_Reduce(&curr_pi, &pi, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); } // Print out the final value and error printf("calculated Pi = %.16frn", pi); printf("Relative Error = %.16frn", fabs(pi - M_PI)); //Wrap up MPI MPI_Finalize(); return 0; } 17/34
- 18. Set allnode IP addresses as static in ◦ Sudo nano /etc/network/interfaces (edit on all nodes) ◦ This step differs between wired and wireless ◦ For wired enter a static etho address ◦ For wireless enter a static address using wlan0 Set all hostnames to now static IP’s ◦ Sudo nano /etc/hosts (edit on all nodes ) ◦ Add in the hostnames and addresses, for example: ◦ rpi0 192.168.0._ ◦ rpi1 192.168.0._ Now we can ssh from one pi to another without having to type IP addresses 18/34
- 19. Setting upthe wireless connection was essentially the same as setting up the wired connection We assigned the ip addresses onto the wireless router Then we went in to the etc/network/hosts and added the new ips with hostnames Added at the bottom of /etc/network/interfaces: iface TP-LINK_7236 inet static ◦ Address 192._._._ ◦ Netmask 255.255.255.0 ◦ Gateway 192._._._ 19/34
- 20. Next acommon user was created on all nodes to allow the nodes to communicate with out the need for repeated password entry ◦ Sudo useradd –m –u 2345 mpiu Next the nodes were mounted onto the master node ◦ Sudo mkdir /mirror //makes the directory ◦ Sudo chown mpiu:mpiu /mirror/ //changes ownership ◦ Sudo service rpcbind start ◦ Sudo update-rc.d rpcbind enable 20/34
- 21. Sudo nano/etc/exports ◦ Line added at bottom of file: ◦ /mirror 192.168.0.0/24(rw,sync) ◦ This line allows all ip addresses from 192.168.0.0 – 192.168.0.255 to be used by this system ◦ This is a possible point of concern when it comes to wireless communication Next nfs server reset and ssh from rpi0->rpi1 Same thing done to rpi1 Then “$~sudo mount rpi0:/mirror” actually mounts the node These steps repeated for all slave nodes 21/34
- 22. SSH Keysgenerated using ◦ Ssh-keygen –t rsa ◦ A passphrase is recommended ◦ A bitmap of random characters was then generated as the key Next key is copied to slave nodes using: ◦ Ssh-copy-id mpiu@rpi1 “keychain logic” added to file .bashrc 22/34 Photo courtesy visualgdb.com
- 23. Log inas mpiu on master node using Su – mpiu Switch to the /mirror/code/ directory which holds the mpi test programs using “cd” Mpicc calc_pi.c –o calc_pi //this line compiles the program Time mpiexec –n 4 –H rpi0-3 calc_pi //this line executes the program on the master node and distributes it to the nodes via the mounts The output is the solution and the time it took to execute 23/34
- 24. Here youcan see the .c files and the executables in the directory You see the execution of the program with mpiexec 24/34
- 25. Initially wehad assumed the code wasn’t working correctly but this proved to be an incorrect diagnosis The times we were seeing were not making much sense We ran the MPI tests on wired and wireless and we found the processing times to be inconsistent 25/34
- 26. This ledus to determine we had an issue with the mounts on the nodes The main issue was that the nodes wouldn’t read the mirrored programs off the master We are still currently in the processes of improving the design and graphically interpreting the data 26/34
- 27. Wired vsWireless performance ◦ Test the processing performance of cluster when: Hard wired to router Using dongles for each node to communicate wirelessly Use wireshark to observe packet latency between nodes Computational benchmark tests ◦ Using benchmark software to observe total processing power across all pi’s ◦ Using complicated program as test material to solve with cluster Graphical performance info Implementation practical applications Active Cooling onto the Pi’s ◦ Adding fans to final case design 27/34
- 28.
- 29. 29/34 Part Price perItem 4Pi's Quantity Total (4) Link Micro SD's 3.28 6 19.68 http://www.newegg.com/Product/Product.a Micro USB's 4.69 4 18.76 http://www.amazon.com/AmazonBasics-Mic Ethernet cables 0.82 4 3.28 http://www.newegg.com/Product/Product.a Wifi Dongles 7.99 4 31.96 http://www.amazon.com/Kootek-Raspberry Router (4-8 ports) 33.99 1 33.99 http://www.newegg.com/Product/Product.a Raspberry Pi's 41.6 4 166.4 http://www.amazon.com/Raspberry-Pi-Mod Heat sinks 2.41 4 9.64 http://www.amazon.com/Cooling-Aluminium Dual Router 19.99 1 19.99 http://www.frys.com/product/8445718?site= Fans 3.95 2 7.9 http://www.tannerelectronics.com Makers Space 35 1 35 https://dallasmakerspace.org Power USB 29.99 1 29.99 http://www.bestbuy.com/site/insignia-7-po Total of All Parts 376.59
- 30. Diagnosing themounting issue Wireless and Wired communication working Final equipment list acquired Measuring and sketching layout of case structures for the laser cutter 30/34
- 31. Compare wiredvs wireless performance ◦ Detailed documenting and graphing of test results Continue to debugging and improving the system ◦ Finish debugging the mounting issue Finish first prototype for final case design ◦ Measuring and cutting the structure of the case 31/34
- 32. Wired andwireless connection is complete Debugging nfs and mounting issues ◦ Continuously running performance tests Final case design blueprint is complete 32/34
- 33. http://www.python.org/doc/current/tut/tut.html http://likemagicappears.com/projects/raspberry-pi-cluster/ http://www.zdnet.com/article/build-your-own-supercomputer- out-of-raspberry-pi-boards/ https://Youtu.be/R0Uglgcb5g http://www.newegg.com/ http://www.amazon.com http://anllyquinte.blogspot.com/ http://www.slideshare.net/calcpage2011/mpi4pypdf 33/34
- 34.